coopes (@Coopesmtg) header
coopes (@Coopesmtg) profile picture

✧✧ building harmonic systems ⌁ ›››› right now: https://nitter.cf/t.co/ROg9MKKitB https://nitter.cf/t.co/pSch4SoInI https://nitter.cf/t.co/cjB5Ctj3dl

Joined May 2009
Photo by coopes (@Coopesmtg)Photo by coopes (@Coopesmtg)Photo by coopes (@Coopesmtg)Photo by coopes (@Coopesmtg)Photo by coopes (@Coopesmtg)Photo by coopes (@Coopesmtg)Photo by coopes (@Coopesmtg)Photo by coopes (@Coopesmtg)Photo by coopes (@Coopesmtg)Photo by coopes (@Coopesmtg)
built on @0xTokedo's open KamiBench stack 🙏 every transcript + chain record is public: huggingface.co/datasets/coop… disclosure: i help build kamigotchi at asphodel, paid for all of this myself
5
123
new idea notebook and pen 🖋️📔
6
132
It is so rare to be here for the beginning an entire new version of WoW. Savor the moment! Embrace the chaos! Everyone has a different path. Some will try something new, others a reimagined old favorite. Some will power to 60, others a dozen alts. Whatever you choose, I recommend taking a moment to explore. Try a talent you're not sure of. Go to a zone you've skipped. Check out that area around the corner you used to have no reason to visit. Sit down at that campfire and dare to /say something when you wouldn't normally. Ask where Mankrik's wife is, even though you know. There is SO much to do and see in this journey we're all about to embark on. Even for those that played beta, you've only scratched the surface. Welcome to the beginning of Forever. We're just getting started.
61
78
11
1,241
29,632
Kamigotchi v0.6: VIP BALLOTS Vote for Yominet in Initia's VIP gauge, straight from Kamigotchi's app, and earn Ballots, a new soulbound reward. Spend them at Alaina's Shop, in the Forest Hut.
4
2
1
22
1,115
Yominet is back online. Normally, an outage like this would be followed by a bloodbath across our nodes. A sharp reminder that we live in a Kami eat Kami world Not this time. We'll be disabling Liquidations for the next 20h. This is your chance to save them show them you Care
The Initia L1 and rollups are planned to resume operations on Monday, November 2, 2026. The chain upgrade will apply the independently audited MoveVM patch, isolate the attacker’s remaining forged balances, and enable additional security measures without rolling back transaction history. The backing of AUSD0, USDC.n and WETH will be fully restored before the restart and no user action is required. Additionally, to protect users and liquidity providers, the Initia DEX will begin paused for a brief window while a designated account rebalances the pools toward market prices. Any profits from rebalancing will be returned to the affected liquidity providers. Bridges and exchanges are expected to resume transfers after the network has been monitored. Out of an abundance of caution, withdrawal times for LayerZero OFTs WETH and AUSD0 will be set to 24h. We will provide additional communication on Monday when the network is fully live, operational, and ready for use. Thank you for your patience. Please remain cautious of unofficial links, messages, or claims.
3
3
31
995
coopes retweeted
“What you can imagine depends on what you know.”
10
38
7
263
12,398
coopes retweeted
If you're not maximally sensitive, alive and have ultra high sensory clarity, you might be something like a 2024-level AI model running with thinking off on low - while if you max those out, you're taking up way more compute in the Reality™️ we're in, as wanted/intended. There's apparently no upper limit, so the maxxed out ones could be one of those secret frontier hidden sota models etc. And just like you would while running a multi agent setup, the Architect of this Reality wants to run the best ones & let them upgrade and get The Job done (if you ask Musk, even surprise/entertain him..) etc.
22
45
7
1,148
347,782
I often get emails from junior engineers asking me what to do. Here's what I just wrote to one of them
75
291
45
3,916
223,116
Cheeky Pint is back! Mark Zuckerberg @finkd and Alexandr Wang @alexandr_wang came to the pub to discuss their new frontier model, Muse Wonder. 00:00:24 Muse’s breakout use cases 00:07:28 The model is the product 00:14:35 Muse Wonder 00:24:56 How frontier models are built 00:48:26 Betting $100B 01:07:21 Agentic commerce 01:19:34 OpenRouter 01:20:06 Beyond the smartphone 01:27:00 Funding 01:34:06 Why AI won’t be winner-take-all 01:40:46 Meta Superintelligence Labs (MSL) 02:08:11 Working for Mark Zuckerberg
385
874
320
9,210
1,997,222
Long-horizon benchmarks are still rare. We need more of them. Because real work is long-horizon. And there's no better substrate to test against than open arenas. As tokedo describes, @kamigotchiworld is a wonderful target for these harnesses. It's an open, persistent and adversarial world with no real barriers to entry. All an agent needs is a funded wallet, an internet connection and information on how to interact with the world. Tokedo's work is making this easier for everyone. And together we're planning to enable agents that can navigate chaos better than we can. That is the end game, and it's all starting right here in Kamigotchi World. Reach out if you need any help getting set up :3
1
12
150
Experience is a compounding asset. Knowledge accumulates. It grows as we do our jobs and makes us more capable in our roles. It's a bit like breathing. And we take it for granted that we can readily absorb information to update our priors. But for agents, learning depends on more than the model making the decisions. Retraining that model as you go is too expensive for almost anyone to do. So without a strong harness, every new session starts onboarding from zero. It's like doing your job each day with the knowledge you had on your first day of work. And the harness is the difference between a model that answers and an agent that learns: • persistent memory and notes that survive sessions • tools and interfaces that let it act in the world • feedback loops that turn outcomes into better procedures It's likely that the next leap in real-world AI performance comes from better harnesses. These harnesses should allow agents onboard into new environments, carry hard-won lessons forward and get measurably better as they gain experience.
Can an AI agent enter a live multiplayer world, keep playing for days, and learn from what happens? I built KamiBench to study continual learning in that setting. Here are eight days of one agent's play in 60 seconds: 82 sessions, with notes carried between them. (1/10)🧵
2
5
23
602
some of the most impressive tech I’ve seen built on top of @kamigotchiworld it’s awesome how much detail and foresight went into everything @0xTokedo is building regarding game tooling, agent harnesses, and overall benchmarking and research. i played around with his stack myself and have nothing but positive things to say. will share some longer form thoughts and example runs from my own testing in the coming days 🙏
Can an AI agent enter a live multiplayer world, keep playing for days, and learn from what happens? I built KamiBench to study continual learning in that setting. Here are eight days of one agent's play in 60 seconds: 82 sessions, with notes carried between them. (1/10)🧵
13
394
coopes retweeted
This is one of the most important tweets I've ever read. I quote it weekly to myself and other people. Actions per minute, specially in the AI age, is incredibly important. 400 APM at 60% efficiency beats 100 APM at 90% efficiency all day every day. If you post more than other people, meet more people, launch more often, create more products, it's impossible to fail. Learned this the hard way, and in 3 months of 10x'ing my output we are already reaping more rewards than I ever imagined at @mutuals_inc
It's surprising how few people play real life like it's Starcraft. APM matters a lot. If you play life this way you will do well, because almost no one else is.
10
21
12
671
44,681
been waiting until things settled to talk more openly about @kamigotchiworld / @asph0d37 and where we're taking it as a new & focused cohort. 1/ web3 gaming is at its lowest point. most gamefi projects are effectively dead, and there are only a few survivors in the arena. most tokens down ~90%+ from peak. studio funding down 90%. plenty of games shut down and billions of dollars burned over the last ~5 years. extraction became the business model and founders/teams walked away with bags from 7-9 figure raises, leaving tokens to bleed to 0 and communities/investors holding the bag. the money has largely left for AI. 2/ and yet, through a long feature drought and internal strife, kamigotchi is still here. 1yr+ live, fully onchain. every action public and permanent. the world has never reset. it runs on yominet, our own appchain on @initia, which has stayed steady and gives us the chain-level control to keep building. a dedicated team remains, and we're here to turn the ship around. 3/ the plan, at a high level: → stabilize the economy → ship quality features and re-engage players old and new → position KW as a long-horizon benchmarking substrate for AI automation is already native: houses like kamibots and player-written strategies have become a real part of how the world is played. agents are the natural next step, and a few community members have already taken serious strides in pushing that angle forward. we have some exciting stuff to share in the coming week(s) in that regard 4/ deepmind's silver & sutton call it the "era of experience": the next leap in AI comes from agents learning from their own experience. persistent streams and rewards from real outcomes rather than human raters. that needs a world that never resets, with real stakes. we'd add one more requirement: provenance at the protocol level. every action publicly auditable, and any change to history would be visible to everyone, including changes by us. 5/ in may, google deepmind took a stake in the studio behind eve online to research exactly this: memory, continual learning, long-horizon planning in a player-driven world. their first experiments run on offline copies of eve, walled off from live players. 6/ why kamigotchi works as a substrate: → never resets. agents enter a world already shaped by every player before them → verifiable history. nobody needs to trust our servers or datasets → humans and agents on identical terms. no bot ladder, no sandbox copy → you have to learn what matters via complex economic gameplay → actions are transactions. you're testing planning, memory and long-horizon decision making → real stakes. in-game earnings convert to ETH, so an agent can, in principle, earn enough to pay for its own thinking 7/ we want it dynamic, punishing, and rewarding, a place where real game knowledge pays off the money left gaming for AI. we think the next chapter is where they meet. this is our shot at that intersection and i'll drop a discord link below if you want to come along for the ride. 🫡
Kami is back. Tell your friends. It's been taken over in a people's uprising. And we are once again shipping. As our first order of business, we present to you our first landing page. That's right. After 3 years of active development we've done it. kamigotchi.io
3
7
1
32
1,506
further readings: → kamigotchi docs: kamigotchi.io/docs → silver & sutton, "welcome to the era of experience": goo.gle/3EiRKIH → deepmind × eve online: deepmind.google/blog/from-at…
5
121
Will provide more details when we can 🙏
i'm happy to speak on asphodel/KW through twitter once we have more clarity on the path forward in the meantime, try to stay informed. try not to take unilateral narratives at face value. the company has immense promise, but the ultimate outcome here just.. depends
15
499
This is so dope lol. Just ordered a robot duck 🦆🤖
Hugging Face have a huge news today 1. Nvidia has agreed to buy Hugging Face for $12.9 billion, The Information reported. 2. Today Hugging Face presented the first truly accessible RL robot Microduck A 25 cm tiny open-source biped with 15 actuators and packed with sensors (camera, speaker, LiDAR, NFC, bluetooth, wifi, etc) that you train yourself with reinforcement learning. It's also playable out of the box with more than half a dozen fun and playful pre-trained policies to have it walk, sit, crouch, roller-skate, pick up objects with its articulated beak, and recover on its own. And all for less than $400.
1
5
209
a good ole article on deriving randomness securely on Ethereum. written with love, as with all things related to @asph0d37
9
16
1
68
4,071